Preview · sales closed. Help, downloads and the customer account are being prepared. Services being prepared
Trust center
Don't take our word for it. Check.
Here is what happens to your voice when you dictate, every connection AURIS can make, and how to check it all yourself.
Updated: 27 September 2026.
Three facts
What we can say, and what you can check.
No telemetry
No usage statistics, no crash reports, no advertising, no sync.
No account
The 15-day trial works without signing up and without a network connection.
Four possible connections
Voice and text are processed on your Mac during dictation. Four network uses are permitted: licence activation, speech-model download, cleanup-model download, and update checks. Updates can be checked manually, when General settings opens, or automatically if you enable the option. None sends dictation audio or text.
When you dictate
Four steps, all on your Mac.
Microphone
AURIS listens only while you hold the key, or in hands-free mode until you stop it.
Transcription on your Mac
With Apple's speech engine on macOS 26 and later; otherwise with whisper.cpp, an open-source engine that runs on your Mac.
Cleanup on your Mac
Rules add punctuation and capitals and remove hesitations. An optional AI cleanup, off by default, also runs on your Mac: Apple's on-device model where available, otherwise a local model you download.
Text at your cursor
Through macOS Accessibility. Without that permission, the text is placed on your clipboard.
What stays on your Mac
What AURIS keeps, and for how long.
| Data | Where | How long |
|---|---|---|
| Audio of a dictation | Temporary file on your Mac | The last successfully processed audio is kept locally for up to five minutes. A recording awaiting crash recovery is kept separately until that recovery is resolved. |
| History | File on your Mac | Your last 20 dictations; can be turned off |
| Glossary and preferences | App preferences on your Mac | Until you delete them |
| License | Signed file on your Mac | As long as the app is installed |
The only connections
Four possible connections. None while you dictate.
| Connection | When | To | What is sent |
|---|---|---|---|
| License activation | You click “Activate”, once per device | Our license server | Your key, your purchase email, a fingerprint of the device that does not reveal its hardware identifier |
| Speech model download | You click “Download” in Preferences | huggingface.co | An ordinary download request; no dictation data |
| Cleanup model download | You click “Download” in Preferences | huggingface.co | An ordinary download request; no dictation data |
| Update check | You click “Check for updates”, you open the General tab of Preferences, or you turn on the automatic check (once a day at most; off by default) | api.github.com, then GitHub | An ordinary request, with no identifier. Nothing is installed without your confirmation |
Any connection shows your IP address to the server it reaches. Buying happens on this website, never inside the app.
Check it yourself
Four ways to check, from two minutes to ten.
Airplane test (2 minutes)
Turn Wi-Fi off, unplug any cable, open Notes and dictate. It works the same.
With LuLu (10 minutes)
Install LuLu, a free and open-source firewall from objective-see.org. Dictate in several apps: no alert names AURIS. As a counter-test, click “Check for updates”: api.github.com appears.
With Little Snitch
Search for “Auris” in the network monitor while you dictate: no connection.
In the Terminal
While you dictate, run
lsof -a -i -n -P -c Auris: no line comes back. Or watchnettop -p Auris.
Seen something else?
Tell us, and we will publish the answer.
If you observe a connection that is not in the list above, or find a security issue in the app or on this site, write to us in English, French or Italian. We acknowledge within 2 business days, and we publish our conclusion on this page.
Under the hood
The engines and models AURIS uses.
Apple's engines
Speech recognition (SpeechAnalyzer) and the on-device language model (Foundation Models) are part of macOS 26. They run on your Mac.
whisper.cpp and Whisper
The speech engine used on earlier macOS versions and for Dutch and Polish. Open source, MIT license.
llama.cpp and Qwen
Used only for the optional AI cleanup when Apple's model is not available. llama.cpp is MIT-licensed; each Qwen model has its own license.
This website
The same rule on our website.
This site sets no tracking cookies and loads no advertising scripts. Payment, when it opens, will be handled by Paddle, our authorized reseller, and its script will load only when you click “Buy”.
Read our privacy policy for the details.
Stay informed
Sales are closed. Join the list to hear when AURIS 1 becomes available.
Mac with Apple silicon, macOS 13 or later.